Twitter is another social networking phenomenon, though in an entirely different way than Facebook is. Where Facebook has privacy concerns because many Facebook users expect what they say to be private and don’t always realize that there are a lot of controls over how your information is shared, Twitter is almost entirely public. When I send a message out on Twitter, called a tweet, it is public. It’s designed to be public. It’s meant to be seen and read by anyone who cares to go looking for it. Because of this, Twitter messages are being archived at the Library of Congress. As of January 2013, there were over 100,000 GB worth of tweets being stored, which, in case you are counting, is over 100 TB. When you consider that each tweet has something under 140 characters in it and each character is generally one byte, we are talking about more than 100,000 billion tweets or, more specifically, as long as you can wrap your head around the number, more than 100 trillion tweets. Let me put this into a different perspective. That’s over 100,000,000,000,000 tweets. And counting, by the way, at a clip of roughly 400 million per day.

Twitter provides a way for people to share thoughts, comments, ideas, information, pointers to documents, and Web sites, as well as pictures. You may think about Facebook as an intimate gathering of your friends around a campfire sharing stories. Twitter, by comparison, is an enormous megaphone on top of a very large building, where people go to shout just about anything they feel like and anyone who cares to listen, can.

The difference between Facebook friends and Twitter followers is that becoming a Facebook friend requires the approval of the person you are requesting the connection with. Twitter, on the other hand, is a very open social network where all you have to do is follow someone and you get to see everything they have to say. This does bring up the issue of being careful what you say on Twitter because if you say it, people will see it. The same is true of all other social media, of course, since it’s easy enough to reshare something someone has posted, but with Twitter it’s not even about resharing, called retweeting, it’s simply the fact that everything you say is public and can be seen by someone interested enough to go searching for it.

Not many years ago, blogging was a popular activity, and the best way to keep up with a lot of blogs was to use Really Simple Syndication (RSS), which was a way of being notified when a new page or blog post got put up. You would get a sample of it through RSS and then you could go to the blog site and read the full page. The world seems like it continues to be more fast paced and sometimes, 140 characters is enough to express a sentiment that could be drawn out to several pages. So, we get microblogging with Twitter where people can express opinions just as they can on a blogging site. We also have a place that acts like an RSS reader. This is one of the things I use Twitter for, actually. It gives me a quick heads-up about things that I may want to look into in more detail.

Twitter can also be used for audience interaction. As I was digging through Twitter and old tweets, I ran across one that was simply the hashtag #dunkgirl. In Times Square, about a year ago, there was a large video screen that had an audience interaction component to it. The screen showed video of people in Times Square standing in front of and underneath it. Superimposed on that was a graphical component, including a large ball that would bounce around based on interaction with the audience. If you saw the ball approaching you on the screen, you could strike out at it in some way and it would bounce off in another direction. In addition to that, there was a short segment where a man and a woman appeared on screen and you were requested to vote on which one you would like to see in a dunk tank. In order to do that, you sent out a tweet with the appropriate hashtag. This is one example of an audience interaction component with Twitter. There are a number of them, including on news shows where tweets may be put onto a scroll or they may use a particular tweet to indicate a preference for one thing or another, much like a vote. This audience interaction increases engagement, and Twitter has become a great way of engaging with customers and audience members.

The engagement level is so high, actually, that people often simply post without thinking and this can be a significant challenge. You may recall Anthony Weiner and all the furor over various things he posted to Twitter. When it comes to Twitter, no matter whether you believe it’s private or not, messages can be permanent. Including the consequences for the posts. Companies may look into your social networking activity before hiring you. And, as in the case with Weiner, you may lose out on something really big if you aren’t careful about what you do.

R is a convenient tool for analyzing text content to discover emergent issues and controversies in diverse corpora. In this case study, I investigate the use of Twitter at a major conference of professional and academic anthropologists. Using R I identify the demographics of the community, the structure of the community of Twitter-using anthropologists, and the topics that dominate the Twitter messages. I describe a series of statistical methods for handling a large corpus of Twitter messages that might otherwise be impractical to analyze. A key finding is that the transformative effect of Twitter in academia is to easily enable the spontaneous formation of information-sharing communities bound by an interest in an event or topic.

To obtain raw data for this study, I searched the Twitter Web site (cf. Gentry, 2011) and downloaded 1500 messages that had been labeled by each message’s author as relevant to the 109th Annual Meeting of the AAA (1500 messages is the maximum number of messages that the Twitter application programming interface (API) allows to download at one time). Authors of Twitter messages frequently use a shared system of notation for identifying the subject of their messages where a hash symbol is placed before the topic word or phrase (Kwak et al., 2010). In this case, the #aaa2011 hashtag was the subject identifier, so I extracted all messages containing this hashtag as follows

In this sample, there are 307 authors whose messages span from 11:00 am EST on 17 until 6:00 pm EST on 20 November 2011. Although the #AAA2011 hashtag was in use in the weeks leading up to the meeting and continued to be used after the meeting concluded, I chose to limit the sample to those produced only during the course of the meeting. There are two reasons for this strategy. First, the scope of this study is limited to a synchronic analysis of Twitter use at the meeting as a time of intensive intellectual and social activity among anthropologists. This is a sampling strategy that has become standard in research on Twitter use in academic and professional contexts because it is a time when people are highly active on Twitter (Ebner, 2009; Ebner and Reinhardt, 2009; Ebner et al., 2010; Letierce et al., 2010a,b; Reinhardt et al., 2009). Second, the Twitter Web site is not explicit about how it makes messages publically available, so it is not always clear if Twitter reveals only a sample of messages matching the hashtag or the entire set of matching messages. During repeated sampling of the Twitter archives, I found I could only obtain a reproducible sample of messages for the period of the meeting, excluding the first day. For the days before the meeting, I was not confident that the archive was making available all the relevant messages. Furthermore, the number of messages in each sample declined with increased time after the event to the point where a few months after the event there were no Twitter messages with the #AAA2011 hashtag. This limitation unfortunately excludes the possibility of tracking which issues generated discussion beyond the meeting and whether Twitter posts during the meeting could predict the staying power of particular topics (cf. Ebner and Reinhardt, 2009; Reinhardt et al., 2009). The unpredictable nature of the results obtained from the Twitter Web site necessitated the exclusion of days for which I could not obtain a reproducible number of messages and more broadly is a serious limitation on the reproducibility of analyses of Twitter corpora.

The use of social media has been growing. Twitter, in particular, due to its simplicity, is a popular platform where users tend to publish their emotional states, ideas and other experiences, almost in real-time. This creates vast amounts of user-generated content, that can cover many topics, and that have attracted the attention of many researchers.

Asur and Huberman (2010) demonstrate how social media, in particular Twitter, can be use to predict real-world outcomes. The study focuses on the prediction of box-office revenues for out-coming movies. In the end, they conclude that a simple model that senses tweets on a particular topic can outperform some market-based predictors, therefore proving the forecasting power of social media. A similar study has been carried on by Tumasjan et al. (2010), focusing in the predictions of Elections. They conclude that the mere number of tweets reflect voter preferences and comes close to election polls, and that the tweets are not only about spreading political opinions, but also to discuss these opinions with other users.

The finding of high-quality content in social media is studied by Agichtein et al. (2008). In the chapter, they investigate methods that allow the identification of high-quality content automatically by exploiting community feedback, such as links between items and explicit quality ratings from members of the community.

Although all the Twitter messages used in this study were publically available at the time the sample was collected, Twitter-users can hide all of their messages at any time, so for the rest of the analysis I have anonymized individual authors here to preserve their confidentiality. The authors include individual and corporate authors (such as the AAA, The Society for the Anthropology of Food and Nutrition, and Wiley-Blackwell). About half of all individual authors in the sample use pseudonyms. The degree of anonymity of the pseudonyms varied greatly. Some authors used a cryptic username unique to their Twitter account with no implied biographical information, giving absolute anonymity to the author. Some used a pseudonym on Twitter that was linked to their physical world self elsewhere on the Internet. Others used a username that could not be linked to a specific physical person, but implied a gender, academic status (e.g., graduate student, postdoctoral scholar, etc.), scholarly interests (e.g., bioanthropology, archeology, or medical anthropology) or some combination of the three.

The use of real names by some of the authors is notable because it links their professional identities as scholars to their authorship of their Twitter messages, giving them ownership of and accountability for their messages. This indicates a use of Twitter by some anthropologists as instrument of professional communication and makes these users visible as the informal public faces of the discipline to Twitter-users. The anonymity preferred by other anthropologists using Twitter is an indication of the heterogeneity of the Twitter-using community and the existence of individuals who prefer to maintain varying degrees of separation between their identity as a Twitter author and other dimensions of their identity (e.g., as a scholar or a student).

Of the 233 authors with a gender-identifying Twitter username (i.e., excluding unrevealing and ambiguous usernames), 128 (55%) identified as female and 104 (45%) as male. Among the 128 authors who provided enough information to determine their academic status, about half of these are graduate students (66, 49%). The next most represented group is faculty at the rank of assistant professors (23, 17%) followed by people with sessional, fixed term appointments, or nontenure track teaching appointments (14, 10%). The remainder is made up of associate professors (11, 8%), full professors (9, 7%), community college faculty (6, 5%), postdoctoral scholars (5, 3%), and undergraduates (2, 1%). In terms of academic status, it seems reasonable to conclude that more junior members of the discipline are most frequently represented on Twitter. This subset may be analogous to Prensky’s (2001) “digital natives” or people whose upbringing was immersed in information and communication technologies, although the presence of more senior academics suggests a mixed group with a range of exposures to technology. Although specific ages for the authors are unavailable for this sample, the relatively small proportion of full professors relative to assistant professors and graduate students suggests that younger scholars are more often users of this form of virtual communication than older ones.

Digital objects, the basic elements of social media systems, vary in size and type. Twitter posts (i.e., tweets) are limited to 140 characters, whereas email messages are typically a few lines to a few paragraphs in length. This difference in size produces dramatically different patterns of interaction. Although size limits of instant messaging are not typically enforced, design choices such as the size of the text box and messaging window promote brevity. Meanwhile, MediaWiki (the wiki platform used by Wikipedia) supports six levels of headers and automatically generates a table of contents, making it relatively easy to create large pages.

The various types of digital objects supported by social media tools are another way to understand the similarities and differences among them. Social media systems have often evolved around a distinct type of digital object: videos at YouTube, photos at Flickr, bookmarks (i.e., web site URLs) at Delicious, books at Amazon, music or podcasts at iTunes, TV shows at Hulu, people at Facebook, tweets at Twitter, messages at discussion forums or email lists, pages at Wikipedia, products at eBay, presentations at SlideShare, 3D objects in Second Life, and career professionals at LinkedIn. Each of these provides different levels and mechanisms of engagement. For example, virtual worlds more closely model embodied physical interactions, where avatars can convey meaning through proximity and orientation [4]. They also introduce many of the burdens of face-to-face interaction, demanding attention to successfully puppet the avatar in interactions with other partners [5]. Although these differences may relate to the type of media (e.g., video, audio, text, 3D model), there are further distinctions within each type. Wikis support structured text elements like tables and bullets, whereas email does not. Some virtual worlds like Webkinz use cartoon characters, while multiplayer games like World of Warcraft include realistic-looking creatures. Of course some social media systems like Facebook include many basic elements: profile pages, wall posts, personal messages, applications, instant messages, notes, groups, photos, tags, status updates, and so on. Wikipedia has user pages, talk pages, articles, edits, categories, and so forth. Even in these systems, identifying the basic elements of the system is important because they are the building blocks of the interactions. They are also the building blocks of networks when they are connected together or exchanged, as you will learn throughout this book.

Online social networks, such as Facebook and Twitter, have become a common part of many people’s lives. It is estimated that Facebook will break the billion user barrier in 2012, and Twitter already claims to have over 100 million users (Head, 2011). Logging on to a personal Facebook page, or leaving a comment on the ‘wall’ of a friend, or of family members, is a daily activity for many people, especially those of the younger generations. Essentially, the use of social media has become so pervasive that it is not unusual to see Twitter messages, links to Facebook pages and blogs displayed during live television broadcasts. In addition, computer-mediated engagement is so prevalent that we are now entering what has been described as ‘the era of social networking, collective intelligence, participation, collaborative creation, and borderless distribution’ (Warr, 2008:1).

As has become apparent, the use of social media has added a new dimension to the old adage: ‘Who you know is often more important than what you know.’ Also, networking with current and prospective clients, colleagues and stakeholders via Internet-based Web 2.0 applications is rapidly becoming a business tool of choice. For instance, Web-based applications that support the creation and exchange of usergenerated content have become an integral aspect of the way business is conducted, and how professional contacts are created and maintained. As Kinzey (2010) pointed out, the social capital accrued through the use of social media can enhance business activities, as well as the public reputation of individuals, groups and organisations.

Universities have also been enthusiastic adopters of Facebook and other social media tools. In this educational environment, social media is used to reach current, former and prospective students, plus engage and inform the broader community of university activity. The use of social media is also making inroads into the traditional curricula of many disciplines. Consequently, and rapidly, it is becoming an integral aspect of the business and commerce curricula in many higher education institutions.

Social networks like Twitter and Facebook provide open platform where people share their text responses. Generally, this activity is termed as text postings. These postings carry information and can help in certain specific domains for its betterment. Text-based classification system can utilize such postings in view to produce summaries. Textual content over social media belongs to either unstructured or semi-structured category. The emergence growth of Web 3.0 online information sources has grownup enormously. The researcher's tried to utilize these sources for identifying meaningful pattern to improve the convenience of human being. Therefore, some automatic tools can help to analyze such large collection of text data. The increasing volume of document demands the relation with some concrete category in a mutually exclusive manner. To address this problem it is found that text categories may belong to many labels. For example the posting related to health might be related to other topics like-disease, diagnosis process, or education. In real world scenario text content may have multiple labels. So, multi-label classification algorithm can suggest possible labels for a given text document. Now days, researchers are attracted in dealing with text analysis to get some valuable insights as a result. Generally, multi-label classification is critical, as it belong to exponential number of possible labels. Text-mining is also a domain of research where only textual data is considered for analysis. Short text messages (like: Twitter data) attracted most of the researchers because of its simplicity, verbosity and character length.

The use of social media (Twitter) postings related to healthcare has been modeled by researchers in many ways. Authors in Ref. [1,8] proposed a dynamic real-time architecture to track early disease outbreak prediction. Support Vector Machine (SVM) based system achieved 88% accuracy in terms of performance. Authors did not raise multi-label context of texts.

Authors in Ref. [2] modeled the problem to overcome the 1–3 week delay of manually aggregating and processing the clinical information. The study proposed a real-time influenza tracking model based on auto regression and Google's search data. The results showed superior approach in terms of accuracy. The study presented in Ref. [3] is very similar to Ref. [2], here author hypothesize the relation between seasonal influenza epidemics as a major public health concern for respiratory illness. The approach uses health queries to detect influenza spread areas.

Authors in Ref. [4,5] presented a real-time flu and cancer related surveillance system based on Twitter data. The model tracks of flu and cancer related activities on twitter. The model has aided visual support which showed the results in US disease surveillance map. The system was helpful for early prediction of seasonal disease outbreaks. The study showed disease (flu and cancer) related monitoring distribution in terms of its types, symptoms and treatments.

In the context of health surveillance using social media, paper [6] proposed a website as a platform for health surveillance using twitter (healthtweets.org). The surveillance mines twitter data and showed health trends in three main domains which are temporal, location and maps. Authors in Ref. [7] devised an auto regression model to predict influenza like illness (ILI) level in population and proposed a framework to predict spread of influenza. The authorized United States of America body, center of disease control (CDC) has reported that Flu is highly correlated with ILI cases. The Framework Social Network Enabled Flu Trends (SNEFT) monitors messages with a mention of Flu indicators. The study claimed substantially improved model in predicting ILI cases using social media. Author in Ref. [9] examine twitter streams to track public sentiments in context of H1N1. Authors have used twitter to measure the actual disease activity in the context of health-related events. The study presented in Ref. [10] used a regression model with ridge regularization and showed improved prediction accuracy. The experimentation combines twitter messages and CDC's ILI data.

In machine learning multi-class-based classification is common problem. However, text content may have multiple labels. Multi-label classification task is not a new concept; it is generalized form of multiclass problem. In multi-label context, study proposed in paper [11,29] demonstrated its application domains like images and videos annotation, music categorization into emotions, functional genomics, etc. Authors exhaustively showed the places where multi-label classification is used widely.

While in paper [12,13] authors have made comprehensive reviews of multi-label learning algorithms which are frequently used by the researchers. Mainly two categories of algorithms are used for classification, transformation and adaptation. Both the papers have discussed on evaluation metrics of multi-label learning. Paper [13] reported the problem of exploiting label correlation by various multi-label learning techniques. Authors in paper [14] presented the most frequently used techniques used to deal with ML problems in pedagogical manner which basically highlighted the similarity measures and differences between techniques.

Work in Ref. [15] signifies the concept of learning objects and validated the findings on four different MLL algorithms. The results were obtained using transformation and adaptation algorithms. Ensemble of Classifier Chains (ECC), Random K-Label sets, Ensemble of Pruned Sets and Multi-label K Nearest Neighbors (MLKNN) were used for experimentation. Authors have used 16 evaluation measures and concluded that ECC perform best among all selected algorithm category. In Ref. [16] authors have used a new ensemble based on set cover problem and demonstrated its effectiveness with constraints, and represented each individual labels with coverage of inter-label correlations. The method performs highly efficient over RAKEL and other ensemble methods.

Authors in Ref. [17] compared 12 MLL methods using 16 evaluation measures over 11 benchmarking dataset and concluded that random forest of predictive clustering trees (RF-PCT) and hierarchy of multi-label classifiers (HOMER) outperforms among all. In Ref. [18] the work addressed learning problem using method multi-label Naive Bayesian (MLNB) which adapts traditional Bayes classifiers and dealt multi-label instances. Two things are incorporated in the proposed algorithm: (1) Feature selection techniques based on PCA and (2) Subset selection technique based on GA. The algorithm helps in finding of appropriate subset of features for prediction. The MLNB suggested that incorporating feature selection is helpful in multi-label learning.

Authors in Ref. [19] proposed Multi-instance multi-label extreme learning machine (MIML-ELM) framework to conceptualize complex objects. The study showed the concept of minimizing high computational cost of SVM when used as a classifier. In Ref. [20] authors described the framework for generating multi-label dataset named as MLDataGen. The result illustrates that hypercube strategy provide better classification than based on hyper-sphere.

In literature there also exists work on hierarchical multi-label classification of social texts [21]. The author proposed the idea to resolve the problem of short text message and its complicated relationships among the classes. The chunk based structural learning framework is used for hierarchical multi-label classification. Study presented in Ref. [22] performed multi-task multi-label (MTML) classification for sentiments and topic classification and tried to overcome the sparsity of microblog data. In the process authors have utilized latent association between tweets sentiment and topics for classification. The proposed MTML algorithm has shown 5% accuracy improvement on sentiments and 12% improvement on topic classification. Multi-label classification for short text messages (Twitter) is a critical task. Twitter data is sparse in nature, it lacks with enough information. Dealing sparsity in multi-label learning is an essential aspect for correct classification.

Health related aspects are very common on social media. People usually share experiences related to their health issues, symptoms and diagnosis related processes frequently. These discussions can be used for identifying health related knowledge from the social networks. In this regard, paper [23] explores the methodology to overcome the limitations of traditional methods, and with the help of n-gram analysis, the study, tried to identify health related insights from social media. The author reported the importance of feature selection algorithm in order to identify best possible features in five heterogeneous feature set. In Ref. [24] the author performs emotion analysis in which belongs to many categories and for this he utilizes graphic emotions, punctuation expressions together using MEC Multi-label emotion classification algorithm, the evaluated results are used for single and multi-label classification using Weibo (a twitter in china) data set.

Author in Ref. [25] have used transformation and adaptation algorithms for ML classification. They have used semi-structured data for ML classification and reported results showed that BR performs better over LP and MLKNN. The author in Ref. [26] proposed a tool to annotate twitter messages related to healthcare, it can also collect tweets. The tool prepares feature set and automatically uses these feature set for relevant text filtering. The authors in Ref. [27] proposed a mechanism to identify incident related information using social media (Twitter). The results are based on situational information and can be used in decision support systems.

Transformation algorithm binary relevance (BR), label power set (LP) and classifiers chain (CC) has used for performance evaluation and SVM is used as a base classifier. Authors have used precision and recall values to compare the results.

With the above reported work, we can say that twitter sparsity play vital role in short text classification. The researchers working in this domain tried to utilize this aspect to model real-time data for text-based expert system. The text-based system can perform surveillance over social media, and can help many ways to raise the quality of real-time decisions. Twitter or any other social media services can be a source for such real-time surveillance. In current multi-label research, researchers did not link text relate preprocessing configurations with ML classifiers. The effectiveness of the configuration can be identified for efficient use of ML Classifiers.

The paper is divided into sections. Section 2 presents the methodology used for multi-label algorithm evaluation. Section 3 explains Disease dataset and Seattle dataset used for experimentation. The measures used for the results evaluation are discussed in Section 4. Section 5 presents architecture for experimenting multi-label algorithms. The data related configuration setups are discussed in Section 6. Section 7 demonstrates the results, its graphical implications and modified ML-KNN algorithm. In last, the study presents conclusion and future work.

Twitter provides news and updates in an engaging way. I use Twitter as a current awareness tool as well as RSS. However, I know some people find there is too much trivia on Twitter for it to be of benefit for keeping up. One effective way to get current awareness updates on particular topics via Twitter is to follow a topic hash tag. Hash tags are assigned by users of Twitter to cover events, conferences and topics, which I described in some detail in Chapter 3.

However, Twitter can become overwhelming and contribute to overload. One way to control this is to use a Twitter client (application software), such as Tweetdeck, to manage and group incoming Twitter messages into lists. Make some lists of the top priority people you follow; others could be the ones you get to when you have spare time. Different Twitter clients include different management tools. The Twitter native interface is adding enhancements all the time and you can now create lists to group people together according to topic or interest. Approach your management of Twitter in the same way as you would with RSS.

Given a stream of messages from Twitter, that is, tweets, our goal is to identify the messages that contain relevant information for traffic characterization. This can, therefore, be seen as a binary text classification problem. There are multiple algorithms for performing text classification, which, given the appropriate training set, achieve relatively high performances. One of the main challenges in this setting is precisely that of generating a representative training set, since percentage relevant messages in the Twitter stream is extremely low (< 0.05%). Thus, manually selecting positive examples among the messages available for training becomes an unfeasibly laborious task.

However, as explained before, there are several official agencies injecting valid traffic messages in the Twittosphere, using robot users. Robot users are quite simple to identify since their messages have very strict formatting. On the other hand, Twitter messages posted by human users tend to be very loose in terms of grammatical structure, and usually contain many spelling mistakes, nonstandard punctuation, and emoticons among other idiosyncrasies. Nevertheless, robot-generated messages provide good examples of vocabulary related to the description of traffic-related events (e.g., names of routes or critical locations in the network), which should be shared by many of the human-generated messages. Therefore, our strategy consists in using robot-sent messages – which are relatively easy to identify – for gathering an initial set of positive examples needed for training a classification model. Given the very small fraction of Twitters related to traffic, a balanced number of negative examples can be randomly chosen from the entire collection of Twitter messages.

We add (1) as additional positive examples to the training set, while (3) are used as negative training examples. The messages in (2) are not included in the expanded training because they can neither be considered negative examples (they are traffic-related), nor they can be considered positive examples (they do not represent the messages we want to capture). Using such expanded training set, we train a second version of the classifier, which is expected to be more robust and to achieve higher precision in identifying relevant traffic messages posted by human users.